"Event loop is closed" error when calling MCP Tools

CrewAI Community Support
1

Hi everyone,

I’m still struggling with an “Event loop is closed” error when my CrewAI agent uses a tool. This tool is a tool on my MCP server.

The Core Problem:

When a CrewAI task invokes a tool (e.g., fetch_notes), I get the following error from CrewAI:

╭────────────────────────────────────────────────────────────────────────────────────── Tool Error ──────────────────────────────────────────────────────────────────────────────────────╮
│                                                                                                                                                                                        │
│  Tool Usage Failed                                                                                                                                                                     │
│  Name: fetch_notes                                                                                                                                                                     │
│  Error: Event loop is closed                                                                                                                                                           │
│                                                                                                                                                                                        │
│                                                                                                                                                                                        │
╰────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯

And in my console logs, I see related messages about an unawaited coroutine, which may be a symptom of the underlying event loop issue:

Exception ignored in: <coroutine object ClientSession.call_tool at 0x14ea2c820>
Traceback (most recent call last):
  File "/opt/homebrew/Cellar/python@3.12/3.12.9/Frameworks/Python.framework/Versions/3.12/lib/python3.12/warnings.py", line 555, in _warn_unawaited_coroutine
    warn(msg, category=RuntimeWarning, stacklevel=2, source=coro)
RuntimeWarning: coroutine 'ClientSession.call_tool' was never awaited

My Setup:

  1. FastAPI Server :

    • The server is a FastAPI application with a fastmcp app mounted.
    • The application running on my FastAPI server calls to my FastMCP server.

    Server main.py:

    from fastapi import FastAPI
from ai.mcp.mcp_server import mcp_app # This is the FastMCP.http_app()
from ai.routes.routes import router # Other app routes

app = FastAPI(lifespan=mcp_app.lifespan)
app.mount("/mcp-server", mcp_app) # Mounting the MCP server
app.include_router(router=router)

    Example Synchronous MCP Tool on Server (mcp.py):

    
@mcp.tool()
def fetch_notes(patientId: str, ctx: Context) -> NoteResponse: 
    token = get_access_token()
    # ... auth checks ...
    url = f"http://EXAMPLE.url"
    headers = {"Authorization": token.token}
    resp = requests.get(url, headers=headers, timeout=10)
    resp.raise_for_status()
    return resp.json()

  2. CrewAI MCP Adapter:

        server_params = {
        "url": "http://localhost:8000/mcp-server/mcp",
        "transport": "streamable-http",
        "headers": {"Authorization": raw.get("token")},
    }
    try:
        with MCPServerAdapter(server_params) as mcp_tools:
            print(f"Available tools: {[tool.name for tool in mcp_tools]}")

Available tools prints fine, listing all the tools the agent has at it’s disposal correctly. However, the error occurs when calling the tools.

Thanks for any insights!

2

I ran into the same issue when using MCP Tool Adapter with STDIO. Any ideas on how to deal with this? They randomly happened, not always.

I encountered an error while trying to use the tool. This was the error: Event loop is closed.
 Tool execute_sql_tool accepts these inputs: Tool Name: execute_sql_tool
Tool Arguments: {'properties': {'sql': {'anyOf': [], 'description': 'The sql to execute.', 'enum': None, 'items': None, 'properties': {}, 'title': '', 'type': 'string'}}, 'required': ['sql'], 'title': 'DynamicModel', 'type': 'object'}
Tool Description: Use this tool to execute sql statement.
3

I have the same issue -Windows machine and venv setup! Even tried to implement the basic add example using STDIO. Always returns this error.

╭─────────────────────────────────────────────────────────────────────────────── Tool Error ───────────────────────────────────────────────────────────────────────────────╮
│ │
│ Tool Usage Failed │
│ Name: add │
│ Error: Event loop is closed │
│ │
│ │
╰──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯

I encountered an error while trying to use the tool. This was the error: Event loop is closed.
Tool add accepts these inputs: Tool Name: add
Tool Arguments: {‘properties’: {‘a’: {‘anyOf’: , ‘description’: ‘’, ‘enum’: None, ‘items’: None, ‘properties’: {}, ‘title’: ‘A’, ‘type’: ‘number’}, ‘b’: {‘anyOf’: , ‘description’: ‘’, ‘enum’: None, ‘items’: None, ‘properties’: {}, ‘title’: ‘B’, ‘type’: ‘number’}}, ‘required’: [‘a’, ‘b’], ‘title’: ‘addArguments’, ‘type’: ‘object’}
Tool Description: Add two numbers (ints or floats)

.venv\Lib\site-packages\crewai\tools\tool_usage.py:265: RuntimeWarning: coroutine ‘ClientSession.call_tool’ was never awaited
return error # type: ignore # No return value expected
RuntimeWarning: Enable tracemalloc to get the object allocation traceback
.venv\Lib\site-packages\crewai\tools\tool_usage.py:269: RuntimeWarning: coroutine ‘ClientSession.call_tool’ was never awaited
return self.use(calling=calling, tool_string=tool_string) # type: ignore # No return value expected
RuntimeWarning: Enable tracemalloc to get the object allocation traceback

4

I managed to get it to work using this approach. Define a Manual connection to start and stop it, then before kicking off the crew start the mcp server and after close it.

e.g. in the crew define two functions.
mcp_server_adapter = MCPServerAdapter(server_params)
mcp_tools =

def start_mcp_adapter():
global mcp_tools
try:
mcp_server_adapter.start()
mcp_tools = mcp_server_adapter.tools
print(f"Available tools: {[tool.name for tool in mcp_tools]}“)
except Exception as e:
print(f"Error starting MCP adapter: {e}”)
mcp_tools =

def stop_mcp_adapter():
if mcp_server_adapter and getattr(mcp_server_adapter, ‘is_connected’, False):
print(“Stopping MCP server connection…”)
mcp_server_adapter.stop()
elif mcp_server_adapter:
print(“MCP server adapter was not connected. No stop needed or start failed.”)

then before kicking off the crew start it and when it’s finished stop it.
start_mcp_adapter()
try:
kickoff crew…
finally:
stop_mcp_adapter()

5

Thanks, this worked for me also.

6

A few weeks ago I added an official way to use MCP Server with CrewBase, check out here.

Let me know if you still facing issues, pls

7

This topic was automatically closed 24 hours after the last reply. New replies are no longer allowed.